Summary of the invention
Invention needs the problem of solution
Under the situation of the coded system of in non-patent literature 2, using, during relevant higher between two sound channels, the performance of ICP is enough.Yet, when low, need the auto adapted filtering coefficient of higher exponent number relevant, according to circumstances, the cost that is used to improve prediction gain is too high.If do not increase the filtering exponent number, then the energy level of predicated error (energy level) might be identical with the energy level of reference signal, and at this moment, ICP is useless.
As far as the quality of voice signal, the low frequency part of frequency band is important in itself.Because the minimum mistake in the low frequency part of decoded voice is significantly lost the whole quality of voice.Because the boundary of the estimated performance of the ICP in the voice coding, during relevant not high between two sound channels, be difficult to reach the satisfactory performance relevant with low frequency part, preferably adopt other coded system.
In non-patent literature 1, in time domain, only the signal of HFS is suitable for ICP.This is a solution to the problems referred to above.Yet, in non-patent literature 1, in the ICP of scrambler, used the input monophonic signal.Comparatively it is desirable to, should use decoded monophonic signal.This is because in the decoding unit side, obtains the stereophonic signal of generation once more through the ICP composite filter, and this ICP composite filter uses the monophonic signal by the monophony decoding unit decodes.Yet; At the monophony scrambler is especially in the audio coding of broadband (more than the 7kHz) during the scrambler of transition coding type such as widely used MDCT transition coding; In order to obtain in coder side, produce the algorithmic delay that some appends at the decode monophonic signal of gained of time domain.
The objective of the invention is to, provide and use the extendible stereo language coding of prediction (ICP) realization between sound channel, and can improve the code device and the decoding device of the estimated performance of the ICP in the stereo language coding.
The scheme of dealing with problems
Code device of the present invention comprises: the monophonic signal generation unit; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained the unit, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained the unit, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum cutting unit, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum cutting unit, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Selected cell; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis unit between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; The said first frequency spectrum cutting unit partly is divided into littler subband part with the intermediate frequency band of said other residual signals; The said second frequency spectrum cutting unit partly is divided into littler subband part with the intermediate frequency band of said monophony residual signals, and said selected cell is to each subband part selection reference signal.
Code device of the present invention comprises: the monophonic signal generation unit; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained the unit, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained the unit, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum cutting unit, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum cutting unit, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Selected cell; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis unit between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; In the length of said reference signal and said echo signal not simultaneously; The forecast analysis unit duplicates a part of or only extraction part of said reference signal and makes the length coupling between said sound channel, and carries out forecast analysis between sound channel.
Code device of the present invention comprises: the monophonic signal generation unit; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained the unit, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained the unit, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum cutting unit, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum cutting unit, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Selected cell; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis unit between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; In the length of said reference signal and said echo signal not simultaneously; Forecast analysis unit extensions or dwindle reference signal and make length coupling between said sound channel, and carry out forecast analysis between sound channel.
Code device of the present invention comprises: the monophonic signal generation unit; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained the unit, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained the unit, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum cutting unit, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum cutting unit, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Selected cell; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis unit between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; In the length of said reference signal and said echo signal not simultaneously; The cycle of said reference signal or said echo signal is asked in the forecast analysis unit between said sound channel, through being that unit duplicates and makes length coupling with the cycle, and carries out forecast analysis between sound channel.
Coding method of the present invention comprises: monophonic signal generates step; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained step, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained step, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum segmentation procedure, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum segmentation procedure, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Select step; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis step between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel, in the said first frequency spectrum segmentation procedure, the intermediate frequency band of said other residual signals partly is divided into littler subband part; In the said second frequency spectrum segmentation procedure; The intermediate frequency band of said monophony residual signals partly is divided into littler subband part, in said selection step, to each subband part selection reference signal.
Coding method of the present invention comprises: monophonic signal generates step; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained step, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained step, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum segmentation procedure, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum segmentation procedure, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Select step; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis step between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; Between said sound channel in the forecast analysis step; In the length of said reference signal and said echo signal not simultaneously, duplicate the part of said reference signal or only extract a part and make the length coupling, and carry out forecast analysis between sound channel.
Coding method of the present invention comprises: monophonic signal generates step; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained step, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained step, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum segmentation procedure, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum segmentation procedure, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Select step; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis step between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; Between said sound channel in the forecast analysis step; In the length of said reference signal and said echo signal not simultaneously, do not expand or dwindle reference signal and make the length coupling, and carry out forecast analysis between sound channel.
Coding method of the present invention comprises: monophonic signal generates step; First sound channel signal of compound stereoscopic acoustical signal and second sound channel signal and generate monophonic signal, and generate other signal as the difference between said first sound channel signal and said second sound channel signal; Other residual signals is obtained step, obtains the other residual signals of conduct to the linear prediction residual difference signal of said other signal; The monophony residual signals is obtained step, obtains the monophony residual signals of conduct to the linear prediction residual difference signal of said monophonic signal; The first frequency spectrum segmentation procedure, the frequency band that said other residual signals is divided into the frequency that is lower than regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; The second frequency spectrum segmentation procedure, the frequency band that said monophony residual signals is divided into the frequency that is lower than said regulation is that low frequency part is the intermediate frequency band part with the frequency band that is higher than the frequency of said regulation; Select step; With the coefficient of frequency of the low frequency part of the coefficient of frequency of the intermediate frequency band part of the coefficient of frequency of the low frequency part of said other residual signals, said monophony residual signals and said monophony residual signals as the reference signal candidate; And with the coefficient of frequency of the intermediate frequency band part of said other residual signals as echo signal; Through checking the relation between said each reference signal candidate and the said echo signal, from said reference signal candidate, select optimal signal as reference signal; And forecast analysis step between sound channel; Carry out forecast analysis between the sound channel of said reference signal and said echo signal and obtain predictive coefficient between sound channel; Between said sound channel in the forecast analysis step, in the length of said reference signal and said echo signal not simultaneously, ask the cycle of said reference signal or said echo signal; Through being that unit duplicates and makes length coupling with the cycle, and carry out forecast analysis between sound channel.
The effect of invention
According to the present invention, as reference signal, use the residual signals of the other signal of reference signal prediction, thereby can improve the estimated performance of the ICP in the stereo language coding through the signal from a plurality of signals, selecting to bring the optimum prediction result.
Embodiment
(embodiment 1)
Below, use description of drawings embodiment 1 of the present invention.In addition, in following explanation, left channel signals, right-channel signals, monophonic signal, other signal are expressed as L, R, M, S respectively, and the regeneration signal of these signals is expressed as L ', R ', M ', S ' respectively.In addition, in following explanation, the lengths table of each frame is shown N, will be expressed as m (f), s (f) respectively the MDCT regional signal (being called coefficient of frequency or MDCT coefficient) of each signal of monophonic signal, other signal.
Fig. 1 is the block scheme of structure of the code device of this embodiment of expression.In code device shown in Figure 1 100, import the stereophonic signal that for example constitutes by each frame by left channel signals in PCM (Pulse Code Modulation, the pulse code modulation (PCM)) form and right-channel signals.
Monophonic signal synthesis unit 101 generates monophonic signal M through synthetic left channel signals L of following formula (1) and right-channel signals R.In addition, monophonic signal synthesis unit 101 uses left channel signals L and right-channel signals R, generates other signal S through following formula (2).Then, monophonic signal synthesis unit 101 outputs to LP analysis/quantifying unit 102 and LP inverse filter 103 with other signal S, and monophonic signal M is outputed to monophony coding unit 104.
In this formula (1), formula (2), n is the time index (time index) in the frame.In addition, the synthetic method that is used to generate monophonic signal is not limited to formula (1).For example, also can use additive methods such as carrying out weighting and method of mixing adaptively to generate monophonic signal.
The other signal S of 102 pairs of LP analysis/quantifying unit carry out analyzing the LP CALCULATION OF PARAMETERS of (linear prediction analysis) and the quantification of the LP parameter that calculates based on LP; And the coded data of the LP parameter that is obtained outputed to Multiplexing Unit 118, the LP coefficient A after will quantizing simultaneously
SOutput to LP inverse filter 103.
LP inverse filter 103 uses LP coefficient A
SOther signal S is carried out the LP liftering, and the residual signals of the other signal that is obtained (below, be called " other residual signals ") Sres is outputed to and adds window unit 105.
104 couples of monophonic signal M of monophony coding unit encode, and the coded data that is obtained is outputed to Multiplexing Unit 118.In addition, monophony coding unit 104 outputs to monophony residual signals Mres and adds window unit 106.In addition, residual signals is also referred to as pumping signal.At most monophony sound encoding device (for example; Based on CELP (Code Excited Linear Prediction; Code Excited Linear Prediction) code device); Perhaps in the code device of the type that comprises the processing that generates LP residual signals or the local residual signals of decoding of quilt, can extract this residual signals.
Add 105 couples of other residual signals Sres of window unit and carry out windowing process (windowing), and it is outputed to MDCT converter unit 107.Add 106 pairs of monophony residual signals of window unit Mres and carry out windowing process, and it is outputed to MDCT converter unit 108.
Other residual signals Sres after 107 pairs of windowing processes of MDCT converter unit carries out the MDCT conversion, and the coefficient of frequency s (f) of the other residual signals that is obtained is outputed to frequency spectrum cutting unit 109.Monophony residual signals Mres after 108 pairs of windowing processes of MDCT converter unit carries out the MDCT conversion, and the coefficient of frequency m (f) of the monophony residual signals that is obtained is outputed to frequency spectrum cutting unit 110.
Frequency spectrum cutting unit 109 is the border with the frequency of regulation, is low frequency part, intermediate frequency band part and HFS with the band segmentation of the coefficient of frequency s (f) of other residual signals, and with the coefficient of frequency s of the low frequency part of other residual signals
L(f) output to low frequency coding unit 111.In addition, frequency spectrum cutting unit 109 partly is divided into littler subband i with the intermediate frequency band of other residual signals, and with the coefficient of frequency s of each subband part of other residual signals
M, i(f) output to icp analysis unit 113,114 and 115.In addition, i is the index of subband and is the integer more than 0.
Frequency spectrum cutting unit 110 is the border with the frequency of regulation, is low frequency part, intermediate frequency band part and HFS with the band segmentation of the coefficient of frequency m (f) of monophony residual signals, and with the coefficient of frequency m of the low frequency part of monophony residual signals
L(f) output to icp analysis unit 115.In addition, frequency spectrum cutting unit 110 partly is divided into littler subband i with the intermediate frequency band of monophony residual signals, and with the coefficient of frequency m of each subband part of monophony residual signals
M, i(f) output to icp analysis unit 114.
The coefficient of frequency s of the low frequency part of 111 pairs of other residual signals of low frequency coding unit
L(f) encode, and the coded data that is obtained is outputed to low frequency decoding unit 112 and Multiplexing Unit 118.
The coded data of the coefficient of frequency of the low frequency part of the other residual signals of 112 pairs of low frequency decoding units is decoded, and with the coefficient of frequency s of the low frequency part of the other residual signals that is obtained
L' (f) output to icp analysis unit 113 and selected cell 116.
Icp analysis unit 113 is made up of sef-adapting filter, and it is with the coefficient of frequency s of the low frequency part of other residual signals
L' (f) as the reference signal candidate, carry out the coefficient of frequency s of each subband part of this reference signal candidate and other residual signals
M, i(f) icp analysis generates an ICP coefficient, and it is outputed to selected cell 116.
Icp analysis unit 114 is made up of sef-adapting filter, and it is with the coefficient of frequency m of each subband part of monophony residual signals
M, i(f), carry out the coefficient of frequency s of each subband part of this reference signal candidate and other residual signals as the reference signal candidate
M, i(f) icp analysis generates the 2nd ICP coefficient, and it is outputed to selected cell 116.
Icp analysis unit 115 is made up of sef-adapting filter, and it is with the coefficient of frequency m of the low frequency part of monophony residual signals
L(f), carry out the coefficient of frequency s of each subband part of this reference signal candidate and other residual signals as the reference signal candidate
M, i(f) icp analysis generates the 3rd ICP coefficient, and it is outputed to selected cell 116.
Selected cell 116 is through the coefficient of frequency s of each subband part of inspection each reference signal candidate and other residual signals
M, i(f) relation between; From the reference signal candidate, select optimal signal as reference signal, and the reference signal ID (Identification) of the reference signal that expression is selected and output to ICP parameter quantification unit 117 corresponding to the ICP coefficient of reference signal.In addition, narrate the detailed explanation of the inner structure of selected cell 116 in the back.
The 117 pairs of ICP coefficients from selected cell 116 outputs in ICP parameter quantification unit quantize, and reference signal ID is encoded.Outputed to Multiplexing Unit 118 to the coded data of the ICP coefficient after quantizing with to the coded data of reference signal ID.
Multiplexing Unit 118 will analyze from LP/coded data of the LP parameter of quantifying unit 102 outputs, from the coded data of the monophonic signal of monophony coding unit 104 outputs, from the coded data of the coefficient of frequency of the low frequency part of the other residual signals of low frequency coding unit 111 outputs and from ICP parameter quantification unit the quantification ICP coefficient coding data and the reference signal ID coded data of 117 outputs multiplexing, and export the bit stream that is obtained.
Fig. 2 is used to explain the structure of the sef-adapting filter that constitutes icp analysis unit 113,114 and 115 and the figure of action.In the figure, H (z) is H (z)=b
0+ b
1(z
-1)+b
2(z
-2)+...+b
k(z
-k), the model (transport function) of its expression sef-adapting filter, for example FIR (Finite Impulse Response, finite impulse response (FIR)) wave filter.Here, k representes the exponent number of auto adapted filtering coefficient, b=[b
0, b
1..., b
k] expression auto adapted filtering coefficient.The input signal (reference signal) of x (n) expression sef-adapting filter, y ' (n) represents the output signal (prediction signal) of sef-adapting filter, and y (n) representes target (target) signal of sef-adapting filter.For example, in icp analysis unit 113, x (n) is equivalent to s
L' (f), y (n) is equivalent to s
M, i(f).
Sef-adapting filter asks the square error (MSE) of prediction signal and echo signal to be minimum auto adapted filtering parameter b=[b according to following formula (3)
0, b
1..., b
k], and with its output.In addition, in formula (3), E{} representes ensemble average computing (ensemble average operation), and k representes the filtering exponent number, and e (n) representes predicated error.
In addition, in the H of Fig. 2 (z), there are a plurality of other structures.Fig. 3 representes one of them.Filter construction shown in Figure 3 is a FIR wave filter in the past.
Fig. 4 is the figure of selection that is used for explaining the reference signal of selected cell 116.In Fig. 4, the number of expression subband is the situation of 2 (i=0,1).In addition, the transverse axis of Fig. 4 is a frequency, and the longitudinal axis is the value of coefficient of frequency (MDCT coefficient), and upside is the frequency band of other residual signals, and downside is the frequency band of monophony residual signals.
At this moment, selected cell 116 is from the coefficient of frequency m of the 0th subband part
M, 0(f), the coefficient of frequency m of the low frequency part of monophony residual signals
L(f), the coefficient of frequency s of the low frequency part of other residual signals
L' (f) in, select the coefficient of frequency s of the 0th subband part of the other residual signals of prediction
M, 0(f) reference signal the time.Likewise, selected cell 116 is from the coefficient of frequency m of the 1st subband part
M, 1(f), the coefficient of frequency m of the low frequency part of monophony residual signals
L(f), the coefficient of frequency s of the low frequency part of other residual signals
L' (f) in, select the coefficient of frequency s of the 1st subband part of the other residual signals of prediction
M, 1(f) reference signal the time.
Fig. 5 is the block scheme of structure of the decoding device of this embodiment of expression.The bit stream that sends from code device shown in Figure 1 100 is received by decoding device shown in Figure 5 500.
Separative element 501 separates the bit stream that decoded device 500 receives; The coded data of LP parameter is outputed to LP parametric solution code element 512; ICP coefficient coding data and reference signal ID coded data are outputed to ICP parametric solution code element 503; The coded data of monophonic signal is outputed to monophony decoding unit 502, and the coded data of the coefficient of frequency of the low frequency part of other residual signals is outputed to low frequency decoding unit 507.
The coded data of 502 pairs of monophonic signals of monophony decoding unit is decoded and is obtained monophonic signal M ' and monophony residual signals M ' res.Monophony decoding unit 502 outputs to the monophony residual signals M ' res that is obtained and adds window unit 504, and monophonic signal M ' is outputed to stereophonic signal computing unit 514.
ICP parametric solution code element 503 pairs of ICP coefficient codings data and reference signal ID coded data are decoded, and ICP coefficient that is obtained and reference signal ID are outputed to ICP synthesis unit 508.
Add 504 pairs of monophony residual signals of window unit M ' res and carry out windowing process, and it is outputed to MDCT converter unit 505.Monophony residual signals M ' res after 505 pairs of windowing processes of MDCT converter unit carries out the MDCT conversion, and the coefficient of frequency m ' of the monophony residual signals that is obtained (f) is outputed to frequency spectrum cutting unit 506.
Frequency spectrum cutting unit 506 is the border with the frequency of regulation, is low frequency part, intermediate frequency band part and HFS with the coefficient of frequency m ' band segmentation (f) of monophony residual signals, and with the coefficient of frequency m ' of the low frequency part of monophony residual signals
L(f) and the coefficient of frequency m ' of intermediate frequency band part
M(f) output to ICP synthesis unit 508.
The coded data of the coefficient of frequency of the low frequency part of the other residual signals of 507 pairs of low frequency decoding units is decoded, and with the coefficient of frequency s of the low frequency part of the other residual signals that is obtained
L' (f) output to ICP synthesis unit 508 and adder unit 509.
ICP synthesis unit 508 is based on the coefficient of frequency m ' of reference signal ID from the low frequency part of monophony residual signals
L(f), the coefficient of frequency m ' of intermediate frequency band part
M(f) or the coefficient of frequency s of the low frequency part of other residual signals
L' (f) in, select one as reference signal.Then, ICP synthesis unit 508 through by following formula (4) expression to quantize the Filtering Processing of ICP coefficient as filter factor, calculate the coefficient of frequency s ' of each subband part of other residual signals
M, i(f), and with it output to adder unit 509.In addition, in formula (4), h (i) is the ICP coefficient, and X (f) is a reference signal, and P is the exponent number of ICP.
Adder unit 509 combines the coefficient of frequency s of the low frequency part of other residual signals
L' (f) with the coefficient of frequency s ' of each subband part of other residual signals
M, i(f), and with the coefficient of frequency s ' of the other residual signals that is obtained (f) output to IMDCT converter unit 510.
The coefficient of frequency s ' of the other residual signals of 510 pairs of IMDCT converter units (f) carries out the IMDCT conversion, and it is outputed to adds window unit 511.The output signal that adds 511 pairs of IMDCT converter units 510 of window unit carries out windowing process, and the other residual signals S ' res that is obtained is outputed to LP synthesis unit 513.
The coded data of 512 pairs of LP parameters of LP parametric solution code element is decoded, and with the LP coefficient A that is obtained
SOutput to LP synthesis unit 513.
LP synthesis unit 513 uses LP coefficient A
SOther residual signals S ' res is carried out the LP synthetic filtering, and obtain other signal S '.
Stereophonic signal computing unit 514 uses monophonic signal M ' and other signal S ', obtains left channel signals L ' and right-channel signals R ' through following formula (5) and formula (6).
L′(n)=M′(n)+S′(n) (5)
R′(n)=M′(n)-S′(n) (6)
Like this, the decoding device 500 of Fig. 5 carries out decoding processing through the signal to the code device 100 of Fig. 1 of receiving, can obtain left channel signals L ' and right-channel signals R '.In addition; As long as the coded data of the coefficient of frequency of the coded data of the coded data of use LP parameter, ICP coefficient coding data, reference signal ID coded data, monophonic signal and the low frequency part of other residual signals forms bit stream, decoding device 500 just can carry out decoding processing.That is to say that as long as the signal that decoding device 500 receives is the signal from the code device that can form such bit stream, it also can need not to be the signal from the code device 100 of the structure of Fig. 1.
The inner structure of selected cell 116 then, at length is described.In this embodiment, expression is based on the situation (first example) of simple crosscorrelation selection reference signal and based on the situation (second example) of prediction gain selection reference signal.
Fig. 6 is the block scheme of the inner structure of the selected cell 116 in expression first example.The coefficient of frequency s of the low frequency part of the other residual signals of selected cell 116 inputs
L' (f), the coefficient of frequency m of each subband part of monophony residual signals
M, i(f), the coefficient of frequency m of the low frequency part of monophony residual signals
L(f), the coefficient of frequency s of each subband part of other residual signals
M, i(f), an ICP coefficient, the 2nd ICP coefficient and the 3rd ICP coefficient.
Coherence check unit 601,602 and 603 calculates simple crosscorrelation through following formula (7) respectively, and will output to simple crosscorrelation comparing unit 604 as the correlation of result of calculation.Here, in formula (7), any signal among X (j) the expression reference signal candidate, in coherence check unit 601, this X (j) is the coefficient of frequency m of each subband part of monophony residual signals
M, i(f); In coherence check unit 602, this X (j) is the coefficient of frequency m of the low frequency part of monophony residual signals
L(f); In coherence check unit 603, this X (j) is the coefficient of frequency s of the low frequency part of other residual signals
L' (f).
Simple crosscorrelation comparing unit 604 selects the highest reference signal candidate of correlations as reference signal, and the reference signal ID of the reference signal that expression is selected outputs to ICP coefficient selecting unit 605.
The ICP coefficient corresponding with reference signal ID selected in ICP coefficient selecting unit 605, and reference signal ID and ICP coefficient are outputed to ICP parameter quantification unit 117.
Fig. 7 is the block scheme of the inner structure of the selected cell 116 in expression second example.The coefficient of frequency s of the low frequency part of the other residual signals of selected cell 116 inputs
L' (f), the coefficient of frequency m of each subband part of monophony residual signals
M, i(f), the coefficient of frequency m of the low frequency part of monophony residual signals
L(f), the coefficient of frequency s of each subband part of other residual signals
M, i(f), an ICP coefficient, the 2nd ICP coefficient and the 3rd ICP coefficient.
ICP synthesis unit 701,702 and 703 calculates each subband coefficient of frequency s ' partly corresponding to the other residual signals of each reference signal through following formula (4)
M, i(f), and respectively output to gain confirmation unit 704,705 and 706.
Gain confirmation unit 704,705 and 706 calculates prediction gain through following formula (8), and it is outputed to prediction gain comparing unit 707.Here, in formula (8), e (n)=s
M, i(f)-s '
M, i(f).Prediction gain Gain in the formula (8) is high more, and then estimated performance is good more.
Prediction gain comparing unit 707 comparison prediction gains selects the highest reference signal candidate of prediction gain as reference signal, and the reference signal ID of the reference signal that expression is selected outputs to ICP coefficient selecting unit 708.
The ICP coefficient corresponding with reference signal ID selected in ICP coefficient selecting unit 708, and reference signal ID and ICP coefficient are outputed to ICP parameter quantification unit 117.
As stated, according to this embodiment, the signal of from a plurality of signals, selecting to bring the optimum prediction result through using the residual signals of the other signal of reference signal prediction, can improve the estimated performance of the ICP in the stereo language coding as reference signal.
In addition, in above-mentioned second example, it is synthetic also can the ICP coefficient after quantizing to be used for ICP.At this moment, replace the ICP coefficient before quantizing, be input to selected cell 116 by the quantification ICP coefficient after the ICP coefficient quantization toleranceization.ICP synthesis unit 701,702 and 703 uses quantification ICP coefficient that other signal is decoded.Based on the comparison prediction gain that predicts the outcome that quantizes the ICP coefficient.In this changes, predict through using the quantification ICP coefficient that in decoding device, uses, can select optimal reference signal.
(embodiment 2)
The situation of ICP coefficient is calculated in embodiment 2 explanations of the present invention after comparing simple crosscorrelation.Fig. 8 is the block scheme of structure of the code device of this embodiment of expression.In addition, in Fig. 8,, and omit its explanation with Fig. 1 to the additional label identical of structure division shared with Fig. 1.Code device shown in Figure 8 800 and code device 100 shown in Figure 1 relatively adopt and have deleted icp analysis unit 113,114 and 115 and selected cell 116, and the structure of having appended selected cell 801 and icp analysis unit 802.
Selected cell 801 is through the coefficient of frequency s of each subband part of inspection each reference signal candidate and other residual signals
M, i(f) relation between selects optimal signal as reference signal from the reference signal candidate, and the reference signal ID of the reference signal that expression is selected outputs to icp analysis unit 802.
Icp analysis unit 802 is made up of sef-adapting filter, and it uses the coefficient of frequency s of each subband part of reference signal and other residual signals
M, i(f) carry out icp analysis, generate the ICP coefficient, and it is outputed to ICP parameter quantification unit 117.
Fig. 9 is the block scheme of the inner structure of expression selected cell 801.The inner structure of the inner structure of selected cell 801 shown in Figure 9 and selected cell 116 shown in Figure 6 compares, and it has deleted ICP coefficient selecting unit 605.
Simple crosscorrelation comparing unit 604 selects the highest reference signal candidate of correlations as reference signal, and the reference signal ID of the reference signal that expression is selected outputs to icp analysis unit 802.
Like this, according to this embodiment, because can relatively calculate the ICP coefficient after the simple crosscorrelation, thus can obtain the effect identical with embodiment 1, and compare with embodiment 1 and can cut down calculated amount.
(embodiment 3)
Embodiment 3 explanation is as the correction ICP to the correction of in the past ICP.Revise ICP and solve the problem that the use length reference signal different with echo signal carried out forecast method.
Figure 10 is the figure of Forecasting Methodology that is used for explaining the correction ICP of this embodiment.In addition, the method with the correction ICP in this embodiment is called " replica method ".In Figure 10, with N
1The length of expression reference signal X (f) (vector) is with N
2The length of expression echo signal.Any signal among X (j) the expression reference signal candidate.
In revising ICP, two situation below considering.
1.N
1=N
2Situation
At this moment, code device uses the ICP of ICP calculating in the past coefficient.In the reference signal of any kind of all this situation possibly take place.
2.N
1<N
2Or N
1>N
2Situation
At this moment, code device is N based on original reference signal X (f) generation length
2New reference signal X
-(f), use new reference signal X
-(f) target of prediction signal, and calculate the ICP coefficient.Then, decoding device uses the method identical with code device to generate X
-(f).When having selected other signal of low frequency or low frequency monophonic signal, produce this situation as reference signal.The length of these signals might be shorter than echo signal, also might be longer than echo signal.
The replica method of this embodiment solves the problem of above-mentioned situation 2.Two following stages are arranged in the replica method.
Step 1: at N
1<N
2Situation under, shown in figure 10, with (the N of the leading section of vector X (f)
2-N
1) point copies to vector X (f) (length N
1) last and generate new vector X
-(f).In addition, at N
1>N
2Situation under, duplicate the initial N of vector X (f)
2Point and generate new vector X
-(f).X (f) is that length is N
2New base vector.
Step 2: use the ICP algorithm, based on vector X
-(f) target of prediction signal s
M, i(f).
Like this, according to the correction ICP of this embodiment,, can both make the subband variable-length of echo signal, and can use the length reference signal different to predict with echo signal regardless of the length of reference signal.That is to say, need not all subbands are divided into the regular length identical with reference signal.Because the low frequency part of frequency band is bigger to the influence that voice quality causes; So the subband of low frequency is divided into shorter length; The subband of the higher frequency that importance is low relatively on the contrary; It is divided into bigger length,, can realizes to expand the raising of the code efficiency in the stereo language coding and the raising of tonequality through being that unit predicts with this dividing frequencyband.
In addition, when selecting the other signal of low frequency, in ICP in the past, need encode and send it to demoder the length reference signal identical with the subband of forecasting object as reference signal.On the other hand, in the correction ICP of this embodiment, the short reference signal of subband that can the utilized bandwidth ratio object is predicted, replaces long reference signal is encoded, and only short reference signal being encoded gets final product.Therefore, the correction ICP of this embodiment can arrive demoder with reference signal transmission with low bit rate.
(embodiment 4)
In embodiment 4, (the N when situation 2 of embodiment 3 is described
1<N
2Or N
1>N
2) alternative method.Forecasting Methodology among the correction ICP of this embodiment is used the value of the point in the short base vector, expands new base vector through interpolation, or base vector is shortened to shorter vector.In addition, the method among the correction ICP of this embodiment is called " expanding/dwindle method ".
The expansion of this embodiment/dwindle has two following stages in the method.
Step 1: at N
1<N
2Situation under, shown in figure 11, through following formula (9) with vector X (f) (length N
1) expand to length N
2Vector X
-(f).
At this moment, in the various method of interpolation such as nearest method of interpolation, linear interpolation method, stereo batten (spline) method of interpolation, Lagrange (Lagrange) method of interpolation any is applicable to X
-(f), ask vector X
-The value of shortcoming point (f).In addition, at N
1>N
2Situation under, through following formula (10) with vector X (f) (length N
1) be reduced into length N
2Vector X
-(f).
Step 2: use the ICP algorithm, based on vector X
-(f) target of prediction signal s
M, i(f).
(embodiment 5)
In embodiment 5, the alternative method that embodiment 3 and 4 are described is (with respect to N
1<N
2Or N
1>N
2The method of situation).Forecasting Methodology among the correction ICP of this embodiment uses long-term forecasting to ask the cycle in reference signal and the echo signal.Through duplicating based on the cycle that is obtained several cycles of original reference signal, generate new reference signal.
Two following stages are arranged in the method for this embodiment.
Step 1: concatenating group calibration signal X (f) and echo signal s
M, i(f) obtain continuous vector X
L(f).Suppose vector X
L(f) there is the cycle in.Minimize through the error e rr that makes following formula (11) and to ask cycle T.In addition, also can wait other computation of Period algorithms to ask cycle T through using correlation method, difference of vibration function (magnitude difference function is with reference to non-patent literature 5).
Here,
At T>min [N
1, N
2] situation under, establish T=min [N
1, N
2].Based on T, be that the signal replication of T once or for several times obtains length N with length from X (f)
2New reference signal X
-(f).
Step 2: use the ICP algorithm, based on vector X
-(f) target of prediction signal s
M, i(f).
In addition, when using the method for this embodiment, need the information transmission of cycle T be arrived decoding device.
In addition; In the explanation of embodiment 3,4 and 5; In the low frequency part of selecting the monophony residual signals during as reference signal; Predict after using any method in the above-mentioned embodiment to generate the reference signal of having expanded the monophony residual signal length, but in addition, the present invention also can generate the reference signal of desired length through the intermediate frequency band that comprises the monophony residual signals.This situation is equivalent to the situation 1 (N that embodiment 3 is put down in writing
1=N
2Situation).
In addition; In embodiment 3,4 and 5; Be divided into subband when predicting in intermediate frequency band, predict successively to the subband of high frequency side, thus under the situation of low frequency part of selecting other residual signals as reference signal from the subband of lower frequency side with other residual signals; Also can use the signal of the subband of the lower frequency side of in advance having accomplished prediction, generate the reference signal of desired length.
More than, embodiment of the present invention has been described.
In ICP; The signal of from a plurality of signals, selecting to bring the optimum prediction result is as reference signal; And use reference signal to predict the residual signals of other signal, so method of the present invention can be called " self-adaptation sound channel prediction (ACP:Adaptive Channel Prediction) ".Through using this ACP of the present invention, can improve the estimated performance of the ICP in the extendible stereo language coding.
In addition; When the monophonic signal encoder/decoder is the transform coder of MDCT conversion etc.; The decoded monophonic signal (or decoded monophony LP residual signals) in MDCT zone can directly obtain from the monophony scrambler in coder side, and can directly obtain from mono decoder at decoder-side.
In addition, the coded system shown in above-mentioned each embodiment is used the other signal (being called the M-S type) of monophonic signal prediction.Also can use a monophonic signal prediction left side or right signal.The action of this moment is in above-mentioned each embodiment, and is except with the other sound channel (L or R are regarded as S) of a left side or R channel replacement, and beyond a left side (or right) sound channel signal encoded, roughly the same with the processing of M-S mode.At this moment; Encode the signal of the opposing party's sound channel (right side or L channel) of sound channel (left side or R channel) of gained in demoder in the coding side; Can use decoded sound channel signal (left side or right-channel signals) and monophonic signal, and calculate according to the mode of following formula (12) and formula (13).In addition, also can with other signal in above-mentioned each embodiment likewise, together two sound channels (L and R) are encoded.
R (n)=2M (n)-L (n) (when coded object is a left side (L) sound channel) (12)
L (n)=2M (n)-R (n) (when coded object is the right side (R) sound channel) (13)
In addition, the present invention also can use said signal the weighted sum signal (with three kinds of signal times with the weighting coefficient of regulation after the signal of addition gained) as the reference signal candidate in above-mentioned each embodiment.In addition, the present invention need not to use the whole of three reference signal candidates, for example also can be only with two kinds of signals of the other signal of the monophonic signal of intermediate frequency band and low frequency as candidate etc.Thus, can reduce the bit number that sends reference signal ID.
In addition, in above-mentioned each embodiment, be the prediction that unit carries out other signal with the frame.This means signal according to the signal estimation intermediate frequency band in the same number of frames on other frequency bands.Replace or in addition, also can use the prediction of interframe.For example, can use over and done with frame to predict current frame signal as the benchmark candidate.
In addition; In above-mentioned each embodiment, explained that the echo signal as forecasting object is to have removed the other signal of the intermediate frequency band of low frequency and high frequency, but be not limited thereto; As echo signal, also can comprise and comprise all signal bands intermediate frequency band and high frequency, except low frequency.And, also can be with all signal bands that comprise low frequency as object.Even still can be that less subband is predicted with any band segmentation of other signal in these cases.Thus, the structure of encoder does not change.
In addition, the present invention also goes for the signal of time domain.For example, also can be from (for example by QMF (Quadrature Mirror Filter, quadrature mirror filter) obtain) several subband signals of time domain the selection reference signal, the centre of prediction time domain (or high) band signal.
In addition, the above illustration that is illustrated as preferred implementation of the present invention, scope of the present invention is not limited thereto.So long as have the system of code device, decoding device, then the present invention can be applicable to any situation.
In addition; Code device of the present invention and decoding device for example can be used as sound encoding device and audio decoding apparatus etc. and are loaded in the communication terminal and base station apparatus of GSM, and communication terminal, base station apparatus and GSM with action effect same as described above can be provided thus.
In addition, though be illustrated as example to use hardware to constitute situation of the present invention here, the present invention also can realize with software.For example, algorithm of the present invention is recorded and narrated, and in internal memory, preserved this program and carry out, thereby can realize and code device of the present invention/decoding device identical functions through signal conditioning package through programming language.
In addition, each functional block that is used for the explanation of above-mentioned embodiment is that LSI realizes as integrated circuit usually.These pieces both can be integrated into a chip individually, also can comprise a part or be integrated into a chip fully.
In addition, though be called LSI here,, also can be called IC, system LSI, super large LSI (Super LSI), especially big LSI (Ultra LSI) etc. according to degree of integration.
In addition, realize that the method for integrated circuit is not limited only to LSI, also can use special circuit or general processor to realize.Also can use can LSI make the back programming FPGA (Field ProgrammableGate Array: field programmable gate array), the perhaps connection of the inner circuit unit of restructural LSI and the reconfigurable processor (Reconfigurable Processor) of setting.
Moreover along with semi-conductive technical progress or other technological appearance of derivation thereupon, if the new technology of the integrated circuit of alternative LSI can occur, this new technology capable of using is carried out the integrated of functional block certainly.Also exist the possibility that is suitable for biotechnology etc.
The spy that on October 31st, 2007 submitted to is willing to that the disclosure of 2007-284622 number the instructions that Japanese patent application comprised, accompanying drawing and specification digest is fully incorporated in the application.
Industrial applicibility
Code device of the present invention and decoding device are suitable for mobile phone, IP phone, video conference etc.